Date:         Fri, 24 Jan 2003 10:46:48 -0600
Reply-To:     Roser Matamala <matamala@ANLEXCH.ANL.GOV>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Roser Matamala <matamala@ANLEXCH.ANL.GOV>
Subject:      nlmixed problem
Content-Type: text/plain; charset="us-ascii"; format=flowed

Hi all,

         Several of you have pointed out that a more specific subject would
be better and adding the code would help to answer the question. Thus here
it is: Hope somebody out there has some experience with nlmixed, I am using
sas version 8.01. The problem is: all the lecture I have done on the matter
tells me that the initial starting numbers for the parms need to be close
to optimum values. I think the values for my parms are as closer as they
can be, but the program keeps saying that "no valid parameter points were
found'.  To give you some history, I have estimated parms by using method
firo and I have scaled up all my parameters. My question to this list is:
has anyone done nlmixed and what does it take for the program to find the
parms?. any help will be considerable appreciate.
Here is the code:

****Covariance Matrix GLM 6.98, -5.02, -3.93, 139.27, -72.45, 57.52
obtained from GLM/df from above;
Proc nlmixed cov data=carbon qpoints=200; *method=firo;

parms b01=35 b11=-2
           b02=63 b12=-11
           b03=15 b13=-2;

         b1=b01+b11*z1+u1;
         b2=b02+b12*z1+u2;
         b3=b03+b13*z1+u3;
         e=exp(-.001*b3*Age);
         pred=b1+b2*(1-e);
model SOC ~ normal(pred,.555);
random u1 u2 u3 ~ normal([0,0,0],[6.98, -5.02, -3.93, 139.27, -72.45,
57.52])subject=tmt out=carbon_glm;
***contrasts;
         contrast 'b01 & b11' b01-b11;
         contrast 'b02 & b12' b02-b12;
         contrast 'b03 & b13' b03-b13;
         title 'nlmixed using the covariance from GLM of nlin of each level';
quit;;

Cheers,


~RM

Roser Matamala, Assistant Ecologist
Environmental Sciences Division
Argonne National Laboratory
9700 S. Cass Avenue
Argonne, IL 60439